Credo Technology Group (NASDAQ: CRDO) has announced a definitive agreement to acquire DustPhotonics, an Israeli fabless semiconductor company specializing in Silicon Photonics Photonic Integrated Circuits (SiPho PICs) for high-speed optical transceivers. The deal values DustPhotonics at $750 million in cash plus approximately 0.92 million Credo shares upfront, with up to 3.21 million additional shares contingent on financial milestones — putting the theoretical ceiling of the deal at approximately $1.3 billion. The transaction is expected to close in Q2 2026 and is projected to be accretive to non-GAAP earnings per share in fiscal 2027. Credo shares rose roughly 13% on the day of the announcement.
The acquisition gives Credo something it has been building toward: vertical integration across the full connectivity stack. The combined company will span SerDes (Serializer/Deserializer), Digital Signal Processing, Silicon Photonics, and system integration — covering both electrical and optical interconnects across AI infrastructure. DustPhotonics, founded in 2017 by Kobi Hasharoni, Amir Garon, and Ben Rubovitch and headquartered in Modi’in, Israel, brings a PIC portfolio rated at 400G, 800G, and 1.6T speeds, with a roadmap extending to 3.2T, and existing design wins at hyperscale cloud customers. The company entered the deal having raised over $100 million in venture capital.
Supply Chain as Strategy
For Credo, the strategic logic is supply chain control as much as it is technology expansion. Silicon Photonics PICs are a foundational component of Credo’s ZeroFlap Optical Transceiver platform. Owning that capability in-house removes external supply dependencies, compresses product development cycles, and opens a pathway to meaningful cost reduction at volume — a factor that becomes decisive as port speeds advance beyond 800G and cluster scale continues to grow. The company expects the combined optical portfolio — ZeroFlap Optical Transceivers, Optical DSPs, and Silicon Photonics products — to generate more than $500 million in optical revenue in fiscal 2027, well above pre-acquisition analyst consensus.
The SiPho PIC market overall is forecast to reach $6 billion by 2030, according to LightCounting and Credo estimates. This is not a niche corner of the semiconductor industry: it is one of the fastest-moving segments in the entire infrastructure stack, driven directly by the insatiable bandwidth and power demands of AI training and inference workloads.
Why Silicon Photonics Now
The context for this deal is a broader inflection point in data center architecture. For decades, electrical interconnects — copper traces, pluggable transceivers, high-speed electrical cables — dominated the inside of data centers. That era is ending. Data center interconnects were consuming nearly 30% of total cluster power in early 2025. At the scale of modern AI training clusters, that is an unsustainable tax. Silicon photonics, which uses light rather than electricity to transmit data between servers and chips, is the structural answer the industry has converged on.
Co-Packaged Optics (CPO) — which integrates optical components directly alongside switching ASICs or processors within the same package — is the most aggressive expression of this shift. CPO eliminates the long electrical traces between the switch ASIC and the optical engine, cutting signal loss, latency, and power consumption simultaneously. NVIDIA’s CPO-based Quantum-X InfiniBand switches began shipping in late 2025, with Spectrum-X Ethernet switches targeting the second half of 2026; NVIDIA’s own data shows CPO reducing interconnect power consumption by up to 3.5x versus traditional pluggable architectures. Broadcom’s 2026 Tomahawk 6-Davisson switch operates at 102.4 Tbps of aggregate capacity using TSMC’s COUPE advanced packaging to stack photonic engines directly onto switching silicon. IDTechEx projects the CPO market will grow at a 37% CAGR from 2026 to exceed $20 billion by 2036.
Near-Port Optics (NPO) — a transitional architecture that places the optical module closer to the switch ASIC without full co-packaging — is the step most hyperscale operators are taking right now, ahead of full CPO deployment. DustPhotonics’ PICs are already in design for both NPO and CPO applications, which means Credo is acquiring not just a current revenue stream but a seat at the table for the architecture that will define data center design for the next decade.
What DustPhotonics Actually Built
DustPhotonics is a fabless company — it designs the chips but does not manufacture them. Its differentiation lies in integrating multiple key optical functions onto a single PIC, rather than assembling them from discrete components. That integration reduces component complexity, improves manufacturing yields, and lowers cost per unit at scale. The portfolio supports both integrated and external laser configurations, giving customers flexibility across different transceiver architectures.
The company assembled a team of approximately seventy engineers with deep photonic integration expertise — a relatively small headcount for the IP and design-win position it achieved. Its products are already deployed in transceivers running inside hyperscale AI clusters. That combination of shipping product, established customer relationships, and a clear roadmap to 3.2T is what justified the valuation.
Credo’s Positioning in a Crowded Field
The silicon photonics landscape is competitive. Coherent Corp., which has been building toward CPO with silicon photonics, InP lasers, and PICs, showed a 6.4T socketed CPO module at OFC 2026. Marvell has integrated CPO technology into its custom XPU architecture. Ayar Labs, Lightmatter, and others are pursuing optical I/O architectures from different angles. TSMC’s manufacturing infrastructure underpins much of the sector.
What Credo is attempting is vertical integration: owning the chip-level building blocks from SerDes through DSP through Silicon Photonics, and packaging that into complete connectivity solutions for hyperscale customers. That is a different bet from being a best-of-breed PIC vendor or a transceiver assembler. It is a systems-level play — one that trades margin per layer for stickiness and control across the full customer engagement. The customer who buys Credo’s AECs, its DSPs, and its optical transceivers is a customer who is significantly harder to displace than one who sources each component from a different vendor.
The risk is execution: integrating a seventy-person Israeli photonics startup into a publicly traded US semiconductor company while simultaneously scaling a new optical revenue stream to $500 million in a single fiscal year is not a conservative program. But the timing, the technology fit, and the existing hyperscale customer overlap between Credo and DustPhotonics reduce the integration risk considerably compared to a cold acquisition.
The Larger Picture
The DustPhotonics acquisition is a signal, not just a transaction. It reflects where the serious money in semiconductor infrastructure has concluded the next five years will be won: not in faster copper, not in incremental transceiver improvements, but in photonic integration that moves the optical engine as close as possible to the compute silicon. The hyperscalers building million-GPU clusters need interconnects that do not become the bottleneck. Silicon photonics is the answer the industry has committed to. Credo, with DustPhotonics absorbed, is now positioned to be one of the few vertically integrated suppliers capable of delivering that answer from chip to cluster.
Leave a Reply